Skip to main content Skip to main navigation

Publikation

Hardware Agnostic Energy Benchmarking For Machine Learning

Timo Laudi; Rolf Drechsler
In: 29. Workshop Methoden und Beschreibungssprachen zur Modellierung und Verifikation von Schaltungen und Systemen (MBMV 2026). ITG/GMM/GI-Workshop "Methoden und Beschreibungssprachen zur Modellierung und Verifikation von Schaltungen und Systemen" (MBMV-2026), March 17-18, Würzburg, Germany, 2026.

Zusammenfassung

While energy consumption of Machine Learning (ML) applications is on the rise, understanding the impact of both hardand software components on runtime efficiency remains a weak point in the research space. We present a hardwareagnostic approach for profiling and characterizing the energy consumption of ML applications leveraging the Open Neural Network Exchange (ONNX) format and ecosystem, demonstrate the portability of the approach over multiple hardware architectures, and highlight its potential to generalize to multiple manufacturers. We further showcase a first proof-ofconcept demonstrating the necessity of clean and comparable training data across all relevant hardware platforms for the development of accurate energy consumption models.